Professional Documents
Culture Documents
1. Introduction
NEMO (Norwegian Economy Model) is Norges Banks
new forecasting and monetary policy analysis model.1
The model is based on international research and model
development over the past 1015 years and has many
features in common with similar models in other central
banks.2 NEMO has been under development since autumn
2004 and has been used previously to analyse specific
developments in the Norwegian economy.
NEMO is based on the assumption that Norway with
a national currency can determine its own inflation level
over time. Therefore, a model requirement is that monetary policy anchors inflation expectations. This means
that monetary policy is decisive in bringing inflation back
to target. In the model, it is assumed that economic agents,
such as households and firms, look ahead when making
decisions concerning consumption, investment, wages
and prices and base these decisions not only on todays
economic policy but also on their expectations concerning future policy. In addition, the model builds on the
experience of the 1970s and 1980s which indicated that
unemployment could not be reduced permanently by
accepting higher inflation. Because of price and wage
stickiness, monetary policy can still influence demand
and hence output and employment in the short and
medium term.
Various agents behaviour is modelled explicitly in
NEMO, based on microeconomic theory. A consistent
theoretical framework makes it easier to interpret relationships and mechanisms in the model in the light of eco-
Thanks to Ida Wolden Bache, Anne Berit Christiansen, Kre Hagelund, Kjersti-Gro Lindqvist, Junior Maih, Bjrn Naug, Kjetil Olsen and Bent Vale
for useful suggestions.
For a more detailed description of the model, see Brubakk et al. (2006).
See, e.g., Adolfsson et al. (2005a,b) Bank of England (2004), Coenen et al. (2007), and Fenton and Murchison (2006).
39
2. The model
NEMO is a modern, Dynamic Stochastic General Equilibrium (DSGE) model which is an extension of real
business cycle theory.4 Dynamic means that the model
solution determines dynamic paths for all endogenous
variables. Developments in the variables will depend,
however, on future random or stochastic shocks which
are not known at the time agents lay their plans. This
means in principle that future realisations of the variables
may be described by a probability distribution, i.e. the
endogenous variables in the model will also be stochastic. General equilibrium implies that market mechanisms
contribute at all times to balancing supply and demand
in all markets in the model. Nevertheless, NEMO provides
a stylised description of the economy, and hence there
are a number of markets, such as the housing market, that
are not represented.
For an introduction to Bayesian econometrics, see for example Koop et al. (2007).
Foreign variables are denoted by a star. Therefore, M* represents foreign imports (of Norwegian produced goods), i.e. our exports.
Technically, this is modelled as a separate intermediate sector which uses the various goods produced as factor inputs.
In the model, petroleum investment is removed as a separate demand component as NEMO is a model for the mainland economy.
7
8
Capital is idiosyncratic but the utilisation rate may change over time.
Technically, we follow Rotemberg (1982) and assume that firms face quadratic adaptation costs. This means that it is profitable for firms to adjust
prices to a certain degree over several periods. A possible interpretation is that a firm risks losing customers if prices are changed to a new level too
rapidly.
40
10
Small letters are used to indicate percentage deviation from long-term equilibrium, i.e. xt = ln(Xt ) ln(X ), where X represents long-term equilibrium
for X, the level variable.
11
In general, we use the term exogenous to describe conditions determined outside the model.
12
We assume the existence of a large number of households and firms, but with the same preferences and technology. The aggregate relationships in the
model will therefore have exactly the same type of function and parameters as similar behavioural relationships for individual agents.
13
To simplify, we let i represent gross coefficients in the model. These are in turn functions of the models structural parameters.
14
The reason is that with a high degree of habit persistence in consumption, households aversion to variations in consumption increases. Thus, they
demand high returns for holding risky assets.
15
Desired wage means the wage level that would be realised if wages are fully flexible. This wage level is equal to the marginal substitution rate between
consumption and leisure and measures households loss of utility, measured in consumption units, as a result of working one hour extra. Households
aversion to working more will in turn depend on how much they work and consume in the first place.
41
(4) S tw
k
(7) it kt1 = 16 Et(it+1 kt ) + 17(it1 kt2) rrt + 18Ert+1
+ zti
q
q
21 22i Et (pt+i
mct+i ) + 23ztq
(9) tq = t1
i=0
16
This may be interpreted as an endogenous risk premium. The interest margin domestic agents pay for foreign loans depends on total foreign debt. High
debt means higher risk and thus a higher premium.
17
In general, demand shocks will result in a negative correlation between profit margins and the output gap, whereas supply-side shocks lead to a
positive correlation. A key premise underlying these results is an assumption of sticky prices. A positive demand shock will, for example, lead to
higher output and increased production costs. Since we assume that firms can only change prices to a limited degree in the short run, firms will not be
able to fully pass on cost increases to prices. As a result, margins decline.
42
m
*
st+i ) mct+i
] + 28ztm
(11) tm = mt1 26 27i Et[(pt+i
i=0
43
2.3 Quantifying
The model is estimated on data for the period 19812007.
We have used quarterly figures for GDP, consumption,
investment, employment, real wages, inflation, imported
inflation, the real exchange rate and the nominal interest
rate. For the foreign variables in the model we have used
trade-weighted data for GDP, inflation, the nominal interest rate and wage costs, bringing the total number of
variables to 13. We have used a relatively large number
of variables in order to be able to estimate the largest
possible number of parameters. Nevertheless, some
parameters in the model are impossible to estimate, given
the information set. These parameters have been calibrated
on the basis of results from international studies and our
own assessments of the transmission mechanism.
We have used a Bayesian approach to determine the
parameters in the model. In general, this means that we
update our initial estimates of the model parameters on
information derived from the data. Our initial view, or
our priors for the probability distribution of the par
ameters, is based on previous assessments and experience
of the functioning of the economy. The final product is
a probability distribution of the parameters, conditional
on the data observed. The Bayesian framework provides
a basis for quantifying the forecast uncertainty in the
NORGES BANK Economic bulletin 1/2009
Exchange
Rate
Central
Bank
Target
Rate
Market
Rate
Imported
Inflation
Inflation
Expectations
Consumption
Investment
Inflation
Labor Market
Wages
Mark-Ups
18
Channels refer to mechanisms in the Norwegian economy through which monetary policy operates.
19
Another channel is the credit channel for monetary policy, on the assumption of imperfect credit markets which can enhance the effects of monetary
policy. The current version of NEMO disregards this aspect. Another version of the model is currently being developed however. Here, some
households are assumed to be credit-constrained.
20
See for example equation (2) which shows how the current level of consumption depends on real interest rate expectations.
44
Chart
inflation
20012007?
Chart44:What
Whatdrives
drives
inflation
20012007?
Percentage
contribution
Percentagepoints
points
contribution
Chart
side
factors
behind
inflation
20012007.
Chart55:The
Thesupply
supply
side
factors
behind
inflation
20012007.
Percentage
contribution
Percentagepoints
points
contribution
-1
-1
Demand side
-2
-3
2001
Exchange rate
-2
Supply side
Foreign
2002
-3
2003
2004
2005
2006
2007
3. Applications
The model may be used both for identifying underlying
forces in the economy and for forecasting purposes. These
two applications are closely connected. Identifying exogenous forces is decisive to making accurate projections.
The model will also be useful in evaluating the published
projections. The structural framework allows the decomposition of the relative contribution of different shocks
to forecast errors for different variables.
-1
-1
Price-setting
Productivity
-2
2001
Wage-setting
2002
2003
-2
2004
2005
2006
2007
Chart
7Forecast
Forecast
errors
different
models
and horizons
Chart 7:
errors
forfor
different
models
and horizons
Root Mean Squared Error (RMSE) for different models and horizons
3.2 Forecasts
Chart 6 shows the quarterly NEMO-based forecasts for
different horizons over the period Q4 1998 Q4 2006
for four key variables. At each time during this period,
a forecast is made for the next eight quarters. The model
is estimated recursively so that in principle we do not use
more information in the estimates than would have been
available at the time the projections were made.21 All
variables are measured as deviations from their respective
averages.
The GDP growth forecasts indicate that NEMO is fairly
accurate as to underlying developments. The model is
less accurate in projecting short-term
movements. We
23
also see that NEMO would have been of limited use in
predicting the fall in inflation from 2002. The interest
rate forecasts seem to be broadly in line with actual
developments. The model also seems to capture the main
developments in the real exchange rate.
We have compared the NEMO-based forecasts with
forecasts from alternative models recommended in the
literature to evaluate the models forecasting properties:
a VAR (2) with no parameter restrictions, a BVAR (2)
21
22
Vector autoregressive models, both without parameter restrictions (VAR) and with a priori parameter spreads (BVAR), are purely statistical models
that describe developments in a set of variables as multivariate autoregressive processes. An AR model is a univariate autoregressive process. BVAR
models in particular have been found to have very good forecasting properties.
23
This is a measure of forecasting accuracy which is calculated by taking the square root of the mean square forecast error for a given variable (and
horizon).
T+Nh 1
24
Both measures are based on the mean square error (MSE) defined as:
1
MSEM (h) =
Nh
(Y
t=T
t+h
where (Yt+h t+h|t ) is the forecast error for a given horizon h, N h indicates the number of projections with horizon h while is a scaling matrix.
We consider ln|MSEM (h)| (log-determinant) and tr(MSEM (h)) (trace).
46
Upper panel shows ln MSE M (h) and lower panel shows tr MSE M (h) .
4. Conclusion
In this article, we have presented Norges Banks new
macro model for the Norwegian economy. The model
contains the most important channels in the Norwegian
economy through which monetary policy operates. In
developing the model, a clear objective has been to
balance clarity and realism. This means that we have
oversimplified in some areas, for example in modelling
fiscal policy and financial markets. Nevertheless, the
model seems to provide a satisfactory description of the
most important developments in25the Norwegian economy
as presented in the data. It cannot be overlooked, however,
that there were problems when estimating the model on
data from periods with different monetary policy regimes.
The estimated parameters must therefore be interpreted
with caution. The models success will ultimately be
assessed on the basis of its usefulness to Norges Bank in
the formulation of monetary policy.
NEMO is one of several models used in the conduct of
monetary policy. In order to obtain a more accurate
picture of todays situation and of developments over the
next few quarters, we rely heavily on various short-term
models. Such models can capture time series properties
in data and correlations that in many cases feature a fairly
high degree of accuracy without the model relationships
following directly from economic theory. In order to
forecast somewhat further ahead, however, we must gain
more insight into the forces at work and how they are
affecting the economy. Statistical forecasting models will
not be particularly helpful here. A solid basis in economic
theory is required to shed light on causal relationships.
47
NEMOs strength is that it provides a consistent description of the relationship between monetary policy and
cyclical developments that lends itself to economic interpretation. The model formalises to a large extent Norges
Banks view of the key relationships in the Norwegian
economy. Thus, it contributes to a consistent framework
that gives structure to both internal discussion and external communication.
A model always involves simplifications and will never
be able to provide an exhaustive description of reality,
but a good macro model can and should be a framework
for ratiocination. In this way, the model plays an important role in the rationale for interest rate setting. At the
same time, it is important to remember that in the conduct
of monetary policy, we must always supplement modelbased results with discretionary assessments.
5. References
Adolfson, M., S. Lasen, J. Lind and M. Villani (2005a):
The Role of Sticky Prices in an Open Economy DSGE
Model: A Bayesian Investigation, Journal of the European Economic Association, vol. 3, pp. 444457
Adolfson, M., S. Lasen, J. Lind and M. Villani (2005a):
Evaluating an Estimated New Keynesian Small Open
Economy Model, Working Paper 203, Sveriges Riksbank
Bank of England (2004): The new Bank of England
Quarterly Model, Quarterly Bulletin 2, Bank of
England.
Brubakk, L., T. A. Huseb, J. Maih, K. Olsen and M.
stnor (2006): Finding NEMO: Documentation of the
Norwegian economy model, Staff Memo 2006/6,
Norges Bank
Coenen, G., P. McAdam and R. Straub (2007): Tax
Reform and Labour Market Performance in the Euro
Area: A Simulation-Based Analysis Using the New
Area-Wide Model, Working Paper 747, European
Central Bank.
Fenton, P. and S. Murchison (2006): ToTEM: The Bank
of Canadas New Projection and Policy-Analysis
model, Bank of Canada Review, vol. 3, 2006
IMF (2004): GEM: A New International Macroeconomic
Model, Occasional paper 239, IMF
Kim J. (2000): Constructing and Estimating a Realistic
Optimizing Model of Monetary Policy, Journal of
Monetary Economics, 45(2)
King R. G. and S. Rebelo (1999): Resuscitating the Real
Business Cycle, Handbook of Macroeconomics, vol. 1b
Koop, G., D. J. Poirier and J. L. Tobias (2007): Bayesian
Econometric Methods, Cambridge University Press
Rotemberg, J. J. (1982): Sticky Prices in the United
States, Journal of Political Economy, 90 (6)
NORGES BANK Economic bulletin 1/2009